Transform-domain Wyner-Ziv Codec for Video

1
Transform-domain Wyner-Ziv Codec for Video
2
Overview

• Low-complexity video encoding
• Transform-domain Wyner-Ziv video codec
• Simulation results

3
Interframe Video Compression

• Current video standards
• Interframe predictive coding for compression
• Encoder is 5-10 times more complex than decoder

4
Low-complexity Video Coding

• Low-complexity encoding, possibly more complex
  decoding
• Applications video sensors for surveillance,
  wireless PC cameras, mobile camera phones

Wyner-Ziv Coding Lossy source coding with decoder
side information
Side Information
Similar work Puri and Ramachandran, Allerton
2002, ICASSP 2003, ICIP 2003
5
Distributed Source Coding
Wyner-Ziv Theorem
Encoder
Decoder
6
Wyner-Ziv Pixel Video Codec
Interframe Decoder
Intraframe Encoder
WZ frames
Slepian-Wolf Codec
Reconstruction
Turbo Encoder
Turbo Decoder
Scalar Quantizer
W
W
Buffer
Request bits
Side information
Y
Interpolation/ Extrapolation
Key frames
Conventional Intraframe decoding
Conventional Intraframe coding
K
K
Aaron, Zhang, and Girod, Asilomar 2002 Aaron,
Rane, Zhang, and Girod, DCC 2003 Aaron, Setton
and Girod, ICIP 2003
7
Wyner-Ziv DCT Video Codec
WZ frames
Decoded WZ frames
W
W
Interframe Decoder
Intraframe Encoder
IDCT
Xk
Xk
qk
qk
Reconstruction

Request bits
bit-plane Mk
Side information
Yk
For each transform band k
DCT
Y
Interpolation/ Extrapolation
Interpolation/ Extrapolation
Key frames
Conventional Intraframe decoding
Conventional Intraframe coding
K
K
8
Quantizers

• Each coefficient band is quantized using a scalar
  quantizer with levels.
  

WZ frame
W

• Mk number of bit planes for kth coefficient
  band
  

4x4 DCT
Xk
For each transform band k

• Combination of quantizers determines the bit
  allocation across bands.

level Quantizer
qk
Sample quantizers Values represent number
quantization levels for coefficient band

• DC band

• Highest freq AC band

9
RCPT-based Slepian-Wolf Codec
Turbo Encoder
Turbo Decoder
Extract bit-planes
qk
qk
Buffer

Request bits
bit-plane Mk
Yk

• Bit planes of coefficients are encoded
  independently but decoded successively
• Rate-compatible punctured turbo code (RCPT)
• Flexibility for varying statistics
• Bit rate controlled by decoder through feedback
• Turbo decoder can perform joint-source channel
  decoding

10
Flexible Side Information

• To generate side information
• Interpolation or extrapolation
• With or without motion-compensation (MC)
• Flexibility in design for decoder
• Laplacian residual model between transform
  coefficients of side information and of current
  frame. Different Laplacian parameter for each
  coefficient band.

Side information
Y
Interpolation/ Extrapolation
Decoded frames
11
Compression Performance

• Mother and Daughter QCIF sequence
• Alternating key frames and Wyner-Ziv frames
  (K-W-K-W)
• Slepian-Wolf codec Rate 4/5 Turbo code
• Side information
• MC-Interpolation
• MC-Extrapolation

12
Compression Performance

• Foreman QCIF sequence
• Alternating key frames and Wyner-Ziv frames
  (K-W-K-W)
• Slepian-Wolf codec Rate 4/5 Turbo code
• Side information
• MC-Interpolation
• MC-Extrapolation

13
Low-Complexity Side Information

• To decrease complexity in generating side
  information
• Average Interpolation
• Previous frame copy
• For low-motion sequences, minimal performance
  degradation
• Transform-domain codec

14
Low-Complexity Side Information
15
Low-Complexity Side Information
Side information Average Interpolation
After Wyner-Ziv Coding 370 kbps
16
H263Intra vs WZ Codec
H263 Intraframe Coding 330 kbps, 32.9 dB
Wyner-Ziv Codec 274 kbps, 39.0 dB
WZ codec Transfor-domain, side information using
MC-Interpolation. Only WZ-frames shown
17
H263Inter vs WZ Codec
H263 I-B-I-B 145 kbps, 40.4 dB
Wyner-Ziv Codec 156 kbps, 37.5 dB
WZ codec Transfor-domain, side information using
MC-Interpolation. Only WZ-frames shown
18
Conclusion

• A new video compression system with intraframe
  encoding but interframe decoding
• Low-complexity encoding for applications such as
  mobile camera phones and video surveillance
  networks, wireless PC cameras
• Transform-domain codec performs up to 2 dB better
  than pixel-domain scheme
• Compared to H.263
• Up to 11 dB better than Intraframe coding
• Performance gap from H.263 interframe coding